Live data from Hacker News

Viewing profile — fastball

fastball

HN member
Joined
Sat, Sep 01, 2012, 1:51 PM UTC
HN karma
12,660
Public activity
6,781 items

About fastball

I used to argue on the internet too much.

Engineering at Known (https://known.com)

Co-founder and tinkerer at Supernotes (https://supernotes.app)

Recent public activity

  1. comment
    Comment #49216707

    But the model is open weight?

  2. comment
    Comment #49200071

    The author didn't say "we (all software engineers)", so the only person including you in his class is you.

  3. comment
    Comment #49176349

    He's had a private Twitter for a while.

  4. comment
    Comment #49173392

    Should've called it Safestral. Also I do like Mistral's seemingly newer strategy of focusing on smaller, more fine-tuned models for various use-cases, presumably the result of thei…

  5. comment
    Comment #49156591

    Arguably the majority of codebases are not available on the public internet

  6. comment
    Comment #49149455

    I remember going from Tiger to Leopard and thinking "this is nice, I like a lot of these changes" and then going from Leopard to Snow Leopard and thinking "wow this is great ". The…

  7. comment
    Comment #49118958

    The thing about data centers and thermoelectric power plants is that you can make them effectively closed cycle, vs agriculture where that is somewhat intractable. Not saying that …

  8. comment
  9. comment
    Comment #49116129

    What I meant by "will they?" was "will they any more than a human already needs to in order to understand other humans?" I don't think this is legibly that different from human beh…

  10. comment
    Comment #49115054

    Will they? This really isn't different from how humans interact with each other. The vast majority of lying is not people being explicitly asked to lie in some form, it is incentiv…

  11. comment
    Comment #49087235

    That is just a single benchmark tho

  12. comment
    Comment #49077952

    Medium vs High? Why? From all the charts I've seen the performance jump is pretty large from med -> high (not as noticeable from high -> xhigh).

  13. comment
    Comment #49074313

    So let's say hypothetically we kept the number of homeless people stable in Oakland, but made them all white. You think all (most? many?) of the people currently complaining about …

  14. comment
    Comment #49073560

    Homelessness is pretty race neutral tho? Especially in Oakland, which is incredibly race-balanced.

  15. comment
    Comment #49064554

    It is actually definitely yes! You can script with JS. Steve Ruiz demonstrates having an LLM do that for you in his intro video to this "offline" version of tldraw: https://www.you…

  16. comment
    Comment #49064138

    The typical meaning of space junk is stuff orbiting around without the ability to control itself and without de-orbiting, therefore presenting a risk/nuisance to other things in or…

  17. comment
    Comment #49063851

    > However, the rapidly increasing numbers of launches to low Earth orbit now threaten to increase the levels of space junk reentering the atmosphere. Isn't it not space junk if it …

  18. comment
    Comment #49061793

    If you want the "agentic future" version of this (sorry), tldraw just relased "tldraw offline"[1] which in some ways has a lot of overlap with HyperCard as well. [1] https://offlin…

  19. comment
    Comment #49060124

    Is it? There are shills and bots all over the internet, and referring to the model as ChatGPT is probably common for some people, but not for software engineers that actually track…

  20. comment
    Comment #49054918

    The fact you called it ChatGPT instead of Codex makes me suspicious of this entire comment.

  21. comment
    Comment #49054703

    Gimbal the floor?

  22. comment
    Comment #49054692

    With so few parameters, I imagine a voice fine-tune might be readily tractable.

  23. comment
    Comment #49043175

    Re-light seems to be one of their biggest remaining obstacles in a lot of these flight tests.

  24. comment
    Comment #49012841

    The problem with the rest of inference is that changes are not trivially correct or incorrect, as they are with the tokenization layer.

  25. comment
    Comment #49012818

    Tokenization is <0.1% of the inference time for the first token in the same way it is <0.1% for the last.